[SCIP] Settings for large continuous bilinear problem

Vladimir V. Voloshinov vladimir.voloshinov at gmail.com
Wed Aug 2 12:43:57 CEST 2017


Dear all,
In combinatorial geometry there is known so called "Spherical Code
problem" AKA "Tammes problem",
https://en.wikipedia.org/wiki/Spherical_code,
http://neilsloane.com/packings.
The problem may be posed as continuous bilinear non-convex Math.
Progr. problem (I used Pyomo to generate AMPL-stub).
I tried to solve it by SCIP 4.0.0 for 3D space up to 14 points (to be
located on 3D-Sphere to maximize minimal "pairwise" distance)
After a number of runs I've seen that SCIP almost at the beginning
finds actually optimal solution and spends all remaining time to prove
its optimality.

But memory became limiting resource here.
Even for 8 3d-points, limit of 28Gb have been exhausted  with gap from
90% till 240% (depending of initial feasible solution, which I put
into AMPL-stub).

Can you recommend any SCIP-options to save memory usage, of course, at
the expense of longer running time?
May be autorestart options may be applied somehow? (I had tried once
but got invariable "dualbound" and killed the process after a long
time).

And one more questions: can you recommend any surveys/reports/articles
devoted to SCIP application for continuous global optimization
(bilinear)?

Sincerely yours,
Vladimir V. Voloshinov.


More information about the Scip mailing list